Characterizing Alzheimer’s Disease and Related Dementia in a Hypertension Population Within the State of Florida Using Electronic Health Record–Based Data

Abstract

Hypertension is a known modifiable risk factor for Alzheimer’s disease and related dementia (ADRD). However, it is unknown how variance in hypertension control, antihypertensive medications, and social determinants of health, such as social deprivation index (SDI), influence the risk of developing ADRD. Validated hypertension computable phenotype algorithms were applied to electronic health record data from the OneFlorida Data Trust (1/1/2013–12/31/2016), to identify apparent treatment‐resistant hypertension (aTRH), and hypertension‐control levels (well‐controlled hypertension, intermediate‐controlled hypertension, uncontrolled hypertension). The primary outcome was a new ADRD diagnosis using validated ICD‐9/10 codes. Multiple adjusted stepwise logistic regression models were used to identify factors associated with ADRD development. ADRD cumulative hazard incidence per hypertension control levels was assessed using the Nelson–Aalen estimator and log‐rank test. A total of 57,273 hypertension patients with 6401 (11%) incident ADRD cases were included in the analysis. The average age was 67 years, with 57% females and 32% identifying as Black or African American. aTRH was a significant ADRD predictor (OR: 1.327, 95% CI: 1.234–1.427), compared to other hypertension phenotypes. aTRH was also significantly associated with a higher incidence of ADRD over time (P < 0.0001). Patients prescribed thiazide diuretics (OR: 0.894, 95% CI: 0.837–0.956) and fixed‐dose combination medications (OR: 0.804, 95% CI: 0.732–0.882) had a lower risk of ADRD. A linear relationship between SDI quartiles and ADRD risk was found. aTRH was significantly associated with the development of ADRD. Our study also highlights the importance of comprehensive hypertension control and socioeconomic interventions in preventing or reducing ADRD risk in hypertension patients.

Study Highlights.

• WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

Hypertension (HTN) is a well‐established modifiable risk factor for Alzheimer’s disease and related dementias (ADRD). However, existing research assessed the effect of HTN based on a single BP measurement or HTN diagnosis, without considering the different levels of HTN control, particularly apparent treatment‐resistant hypertension (aTRH) and its impact on ADRD risk. It is also unknown how different antihypertensives and the social deprivation index (SDI) influence ADRD risk.

• WHAT QUESTION DID THIS STUDY ADDRESS?

This study addressed how different HTN control levels (aTRH, intermediate‐controlled, and uncontrolled HTN), compared with well‐controlled HTN, are associated with an increased risk of developing ADRD. The study also addressed how different antihypertensives and SDI influence ADRD risk among hypertensive patients.

• WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

This study demonstrated that patients with aTRH had a significantly higher risk of developing ADRD, compared to well‐controlled HTN. Additionally, thiazide diuretics and fixed‐dose combination therapy were associated with lower ADRD risks, while a higher SDI score correlated with increased ADRD risk.

• HOW MIGHT THIS CHANGE CLINICAL PHARMACOLOGY OR TRANSLATIONAL SCIENCE?

These findings highlight the importance of identifying and aggressively managing patients with aTRH to mitigate ADRD risk. Clinicians should prioritize aTRH as a distinct high‐risk phenotype, guiding targeted interventions, including tailored antihypertensive treatments and socioeconomic support strategies to reduce ADRD risk.

Alzheimer’s disease and related dementias (ADRD) represent a group of neurological disorders characterized by a progressive decline in cognitive function, affecting memory, thinking, and behavior, primarily in older adults. ¹ The prevalence of ADRD is a growing concern globally, with current estimates that ~6.7 million Americans are living with ADRD, a number expected to double by 2060. ² Currently, available treatments for ADRD primarily manage symptoms rather than halt or reverse the disease progression. ³ Consequently, prevention and risk reduction have become crucial strategies in combating ADRD. ⁴ The 2020 Lancet Commission on Dementia Prevention, Intervention, and Care identified 12 modifiable risk factors that collectively contribute to approximately 40% of dementia cases worldwide. ⁵ Evidence indicates that controlling these risk factors across the life span could potentially delay or prevent about 40% of ADRD cases ⁴ , ⁵ , ⁶ .

Hypertension (HTN) is one of the significant, modifiable risk factors for ADRD ⁷ , ⁸ , ⁹ . With nearly half of the U.S. adult population affected by HTN, the potential impact of effective blood pressure (BP) management on reducing ADRD’s burden is substantial. ¹⁰ , ¹¹ The 2020 Lancet report suggested that maintaining a systolic BP of < 130 mm Hg is an effective strategy for mitigating ADRD risk. ⁵ A recent meta‐analysis of 136 studies showed that antihypertensive use was associated with a 21% reduction in ADRD risk. ¹² Additionally, prior studies have shown that HTN, especially midlife uncontrolled HTN, significantly increases the risk of ADRD. ¹³ However, much of the existing literature on HTN and the risk of ADRD has primarily focused on a single BP measurement, the presence of a HTN diagnosis, or treatment with antihypertensive medications, neglecting the variance of HTN control over time. ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ Such approaches overlook the complexity of HTN, particularly the differential impact that controlled HTN and apparent treatment‐resistant hypertension (aTRH) may have on the development and progression of ADRD.

A HTN diagnosis encompasses all levels of HTN control from controlled HTN to aTRH. ¹⁸ Controlled HTN refers to effectively managed BP with one or two antihypertensive classes. In contrast, aTRH is defined as BP that remains above the target level despite the concurrent use of ≥ 3 antihypertensive classes, or controlled BP with ≥ 4 antihypertensive classes. ¹⁸ The term aTRH is used when pseudoresistance, such as white coat HTN or medication nonadherence cannot be ruled out. ¹⁸ The distinction between controlled HTN and aTRH is crucial, as it reflects varying levels of risk. aTRH is associated with a higher risk of adverse cardiovascular and cerebrovascular outcomes, suggesting a potential parallel risk in the context of ADRD. ¹⁹ , ²⁰ , ²¹ , ²²

Our study aimed to examine how variations in HTN control (aTRH vs. controlled HTN) and the use of antihypertensive medications influence the risk of developing ADRD. To that end, we applied our validated HTN control computable phenotype algorithms to electronic health record (EHR)‐based data from HTN patients within the OneFlorida Data Trust. ²³ We also explored the impact of social determinants of health such as variations in social deprivation index (SDI) and healthcare access within the HTN population on the risk of ADRD.

METHODS

Data source

This was a retrospective observational cohort study performed on EHR‐based data from ambulatory HTN patients, extracted from the OneFlorida Data Trust, from January 1^st, 2012, through March 31^st, 2019. The OneFlorida Data Trust is a centralized, patient‐level research data repository managed at the University of Florida, housing EHR, claims, and death data, on > 21 million individuals, from 15 healthcare partners across the state of Florida. Data are stored in the Patient‐Centered Clinical Research Network (PCORnet) common data model (CDM) to ensure data element standardization. ²⁴ The OneFlorida Data Trust employs a deduplication process via privacy‐preserving entity resolution to link patients’ EHR, claims, and death data. ²⁵ Data were extracted in the PCORnet CDM format from the OneFlorida Data Trust on August 16, 2019, for the following EHR‐based tables: demographic, condition, diagnosis, encounter, procedure, prescribing, and vital. The study was approved by the University of Florida Institutional Review Board, with a full waiver of informed consent for research involving data previously collected for nonresearch purposes.

Patient population

Patients were included in the study if they met the following criteria: age ≥ 50 years, a HTN diagnosis from an ambulatory outpatient visit defined by the International Classification of Diseases, Ninth Revision (ICD‐9: 401.0, 401.1, 401.9) or Tenth Revision (ICD‐10: I10) codes, at least two BP measurements recorded at two separate ambulatory outpatient visits, where systolic and diastolic BP are not null, and at least one prescribing record for antihypertensive medication (Figure S1 ). Patients were excluded from the analysis if they had a pregnancy, secondary HTN, or ADRD diagnosis (Table S1 ) before the index date, or did not meet the criteria for any of the HTN computable phenotypes (detailed below).

HTN computable phenotype algorithms

HTN computable phenotype algorithms that we have previously developed and validated through EHR manual chart review were applied to EHR‐based data from the HTN population, ²³ after data preparation and cleaning, using an index period of January 1, 2013, through December 31^st, 2016 (Figures S1 and S2 ). ²³ RxNorm Concept Unique Identifiers (RxCUI) were used to identify and link antihypertensive medications in the dataset. A medication exposure variable was created to count the number of distinct antihypertensive medication classes that patients were prescribed on a daily basis. ²⁶ Systolic and diastolic BP measurements were assessed during the antihypertensive medication prescribing windows. BP control was defined as BP < 140/90 mm Hg. ²⁷ Current Procedural Codes were used to indicate routine outpatient care, as previously described. ²³ If a patient had multiple BP measurements at a single encounter, the lowest BP measurement was selected to minimize the risk of white coat HTN.

Patients were then classified into HTN control levels. First, patients were considered to have aTRH if they met either of the following criteria: (1) were prescribed four different antihypertensive classes concurrently, with controlled or uncontrolled BP, or (2) were prescribed three different antihypertensive classes concurrently while their systolic BP was ≥ 140 mm Hg or diastolic BP was ≥ 90 mm Hg, at least 1 month after being prescribed the third antihypertensive class. These criteria were required to be met twice, at least 30 days apart, but within a 3‐year window, during the index period (Figure S2 ). ²³ aTRH patients were further divided into controlled aTRH and uncontrolled aTRH based on BP at the index date (i.e., uncontrolled aTRH: systolic BP ≥ 140 mm Hg or diastolic BP ≥ 90 mm Hg, or controlled aTRH: BP < 140/90 mm Hg).

Patients not meeting aTRH criteria were classified into non‐aTRH control levels: well‐controlled HTN, intermediate‐controlled HTN, and uncontrolled HTN. Patients were considered to have well‐controlled HTN if they had controlled BP at > 80% of all outpatient visits over the index period and were only ever prescribed 1–3 different antihypertensive classes simultaneously. Patients were considered to have intermediate‐controlled HTN if they had controlled BP at 50–80% of all outpatient visits over the index period and were only ever prescribed 1–3 different antihypertensive classes simultaneously. Patients were considered to have uncontrolled HTN if they had controlled BP < 50% of all outpatient visits over the index period and were only ever prescribed 1–2 different antihypertensive classes simultaneously (Figure S2 ).

Patients were then grouped into an overall controlled HTN group and an overall uncontrolled HTN group based on their BP control rate. The overall controlled HTN group included well‐controlled and intermediate‐controlled HTN, and controlled aTRH, whereas the overall uncontrolled HTN included uncontrolled HTN and uncontrolled aTRH.

The index date was defined as the first date the patient met the classification criteria for aTRH, or one of the HTN control levels (e.g., the second BP measurement, or 30 days after the anti‐HTN prescription requirement was met) (Figure S2 ).

Patients who did not meet the full criteria for any of the HTN computable phenotypes were patients who were classified into a phenotype group once during the index period, without meeting the criteria for that phenotype twice, at least 30 days apart within the index period, and patients with no antihypertensive prescription exceeding a 30‐day supply during the index period (Figures S1 and S2 ).

SDI and healthcare access score determination

SDI and healthcare access scores were derived based on patient home ZIP codes, which were converted to ZIP code tabulated areas (ZCTAs) using the Uniform Data System Mapper ZIP code to ZCTA crosswalk. ²⁸ SDI and healthcare access were based on the most recent residential ZIP code available in the patient’s records. SDI is a comprehensive measure that incorporates various indicators of social determinants of health, including education, employment, housing, and poverty by ZCTA across the United States. ²⁹ A healthcare access score was calculated by using the two‐step floating catchment area (2SFCA) method to measure the spatial accessibility (drive time) between ZCTA population centers and primary care provider locations. ³⁰ SDI and healthcare access were determined using a previously published method. ³¹ Both SDI and healthcare access were stratified into four quartiles. Quartile 1 in the SDI represents the least social deprivation. In healthcare access, quartile 1 represents the greatest geographical access. Quartile 1 in both measures was chosen as the reference group to reflect the most favorable conditions characterized by minimal social deprivation and maximal geographic access, while quartile 4 would represent the least desirable conditions characterized by high levels of social deprivation and limited geographic access.

Covariates

Comorbidities were defined using ICD‐9/10 codes at or before the index date (Table S1 , Figure S2 ). Drug exposure and average BP metrics were determined at the index date (Figure S2 ). Antihypertensive drug classes included in the analysis were angiotensin‐converting enzyme inhibitors (ACEIs), calcium channel blockers (CCBs), β‐Blockers (BBs), angiotensin receptor blockers (ARBs), thiazide‐like diuretics, aldosterone antagonists, other diuretics (e.g., potassium‐sparing, loop), and other antihypertensives (e.g., vasodilators, renin inhibitors, centrally acting α‐2 agonists, and alpha‐blockers). We also assessed the use of fixed‐dose formulations (Table S2 ).

Follow‐up and outcomes ascertainment

The study outcome was incident ADRD. Patients were followed up after the index date until the diagnosis of ADRD (Figure S2 ). ADRD was captured using ICD9/10 diagnosis codes and included Alzheimer’s disease, vascular dementia, Lewy body dementia, memory loss, dementia, and other dementias using codes that have been previously validated (Table S1 ). ³² Patients who did not have the outcome were censored at their last encounter date during the study period (> 95% of the population had a follow‐up encounter during the study period) or at the end of the study, March 31^st, 2019 (Figure S2 ).

Statistical analysis

Patient characteristics and comorbidities were characterized using descriptive statistics. Categorical variables were reported as frequencies and percentages and compared using the Chi‐square test, while continuous variables were reported as mean with standard deviation and compared using analysis of variance (ANOVA). A univariate logistic regression analysis was performed to analyze the relationship between ADRD and independent variables. Variables significantly associated with ADRD and their odds ratio (OR) and 95% confidence interval (CI) were identified using a backward stepwise logistic regression model, with a cutoff of P‐value < 0.05 for a variable to enter the model and a threshold of a P‐value of < 0.001 to stay in the model. Backward stepwise logistic regression was chosen to allow clinical interpretability of the model and variables. To account for collinearity between HTN control levels, and antihypertensive medications, multiple backward stepwise logistic regression models were developed. Model 1 included significant covariates from the univariate analysis plus overall uncontrolled HTN (overall controlled HTN as a reference group). Model 2 included significant covariates from the univariate analysis plus the HTN control levels (aTRH, intermediate‐controlled HTN, and uncontrolled HTN, with well‐controlled HTN as a reference group). Model 3 added antihypertensive medication classes to Model 2. Similar multiple backward stepwise logistic regression models were performed in subgroups, including aTRH, Black or African American race, Hispanic ethnicity, SDI < 50%, healthcare access < 50%, overall controlled HTN, overall uncontrolled HTN, and in each ADRD subtype. Additional subgroup analyses were performed to examine the risk of aTRH on ADRD by the patient characteristics and comorbidities that were identified from the main analysis using adjusted multivariate logistic regression.

In order to examine the influence of time on ADRD development, Cox proportional hazard regression models were performed to estimate the hazard ratio (HR) of the effect of HTN control levels, and other variables on ADRD. Two models were performed. Model 1 was adjusted for patient demographics, common comorbidities, and risk factors identified by the 2020 Lancet Commission on Dementia Prevention, Intervention, and Care (diabetes, depression, brain injury, hearing loss, obesity, smoking, and alcohol‐related disorders). ⁵ Model 2 was adjusted for variables in model 1 plus the HTN control levels (aTRH, intermediate‐controlled HTN, and uncontrolled HTN, with well‐controlled HTN as a reference group). The cumulative hazard function of ADRD by HTN control levels was assessed using Nelson‐Aalen estimators. The log‐rank test was used to compare the cumulative hazard of ADRD across HTN control levels. Data cleaning, variable creation, and statistical analyses were conducted in SAS version 9.4 (SAS Institute Inc., Cary, North Carolina) and R version 4.2 (R Foundation for Statistical Computing).

RESULTS

HTN control levels, patient characteristics, and ADRD development

741,255 individuals with a HTN diagnosis were identified within the OneFlorida Data Trust between January 2012 and March 2019. Of these, 155,380 patients met the screening criteria during the index period (January 1^st, 2013 to December 31^st, 2016). 98,107 were excluded due to a diagnosis of pregnancy or secondary HTN, ADRD before the index date, or not meeting the criteria for any HTN computable phenotype (Figure S1 ). The final cohort included 57,273 patients for HTN phenotype categorization and outcomes assessment (Figure S1 and S2 ).

Patient demographics and characteristics overall and in each HTN control level are presented in Table 1 . The average age was 67 years, with the majority being female (57%), and 32% identifying as Black or African American (Table 1 ). We identified 18,465 (32.2%) patients with aTRH, 14,752 (25.8%) with well‐controlled HTN, 12,949 (22.6%) with intermediate‐controlled HTN, and 11,107 (19.4%) with uncontrolled HTN (Table 1 ). There were 20,799 patients grouped into the overall uncontrolled HTN group, and 36,474 patients grouped into the overall controlled HTN group (Figure S1 ). The follow‐up times by HTN control level were as follows: 3.2 years for aTRH, 3.2 years for well‐controlled HTN, 3.3 years for intermediate HTN, and 3.1 years for uncontrolled HTN. aTRH was significantly associated with a higher frequency of comorbidities compared to other HTN control levels. Additionally, patients classified as aTRH were more likely to be older, identify as Black or African American, and have a higher BMI (Table 1 ).

Table 1.

Baseline characteristics of the overall HTN population, and HTN phenotypes within OneFlorida Data Trust

Characteristics	Overall HTN (57,273)		Well‐controlled HTN (14,752)		Intermediate‐controlled HTN (12,949)		Uncontrolled HTN (11,107)		aTRH (18,465)		P‐valuea
ADRD	6401	(11%)	1480	(10%)	1367	(11%)	1015	(9%)	2539	(14%)	< 0.0001
Age, mean (SD)	67.6 (10.5)		66.8 (10.2)		67.1 (10.4)		67.3 (10.6)		68.9 (10.5)		< 0.0001
Sex (Female)	32,595	(57%)	8177	(55%)	7551	(58%)	6506	(58%)	10,361	(56%)	< 0.0001
Race											< 0.0001
Black or African American	18,146	(32%)	3158	(21%)	3412	(26%)	3543	(32%)	8033	(44%)
Otherb	3182	(5%)	995	(7%)	738	(6%)	624	(6%)	825	(5%)
White	35,945	(63%)	10,599	(72%)	8799	(68%)	6940	(63%)	9607	(52%)
Ethnicity											0.0020
Hispanic	1866	(3%)	521	(4%)	452	(4%)	363	(3%)	530	(3%)
Body mass index, kg/m2											< 0.0001
< 25.0	12,269	(21%)	3622	(25%)	2935	(23%)	2593	(24%)	3119	(17%)
25.0 to < 30.0	16,761	(29%)	4680	(32%)	3974	(31%)	3275	(30%)	4832	(26%)
≥ 30.0	27,909	(49%)	6415	(44%)	5987	(46%)	5170	(47%)	10,337	(57%)
Smoking											< 0.0001
Current smokers	35,443	(62%)	9058	(62%)	8172	(63%)	7236	(65%)	10,977	(60%)
Former smoker	21,350	(37%)	5606	(37%)	4686	(36%)	3772	(34%)	7286	(39%)
Others	288	(< 1%)	64	(< 1%)	60	(< 1%)	51	(< 1%)	103	(< 1%)
Comorbidities
Diabetes	23,211	(41%)	5388	(37%)	4560	(35%)	3671	(33%)	9592	(52%)	< 0.0001
Dyslipidemia	39,640	(69%)	10,483	(71%)	8815	(68%)	7001	(63%)	13,341	(72%)	< 0.0001
Stroke	2851	(5%)	607	(4%)	537	(4%)	480	(4%)	1227	(7%)	< 0.0001
Chronic kidney disease stage 1–3	8751	(15%)	1773	(12%)	1401	(11%)	1194	(11%)	4383	(24%)	< 0.0001
Chronic kidney disease stage 4–5	1992	(3%)	320	(2%)	285	(2%)	235	(2%)	1152	(6%)	< 0.0001
Myocardial infarction	4715	(8%)	1326	(9%)	785	(6%)	555	(5%)	2049	(11%)	< 0.0001
Angina	6280	(11%)	1635	(11%)	1272	(10%)	874	(8%)	2499	(14%)	< 0.0001
Heart failure with preserved ejection fraction	7454	(13%)	1699	(12%)	923	(7%)	608	(5%)	4224	(23%)	< 0.0001
Coronary artery disease	13,235	(23%)	3587	(24%)	2345	(18%)	1591	(14%)	5712	(31%)	< 0.0001
Atherosclerosis	6191	(11%)	1371	(9%)	1190	(9%)	1014	(9%)	2616	(14%)	< 0.0001
Cardiomegaly	5401	(9%)	1148	(8%)	733	(6%)	638	(6%)	2882	(16%)	< 0.0001
Alcohol‐related disorders	2968	(5%)	715	(5%)	718	(6%)	582	(5%)	953	(5%)	0.0741
Composite of anxiety and depression disorders	19,546	(34%)	5432	(37%)	4674	(36%)	3559	(32%)	5881	(32%)	< 0.0001
Sleep apnea	7706	(13%)	1932	(13%)	1517	(12%)	1121	(10%)	3136	(17%)	< 0.0001
Hearing loss	5953	(10%)	1627	(11%)	1401	(11%)	1135	(10%)	1790	(10%)	0.0003
Brain injury	3781	(7%)	919	(6%)	922	(7%)	742	(7%)	1198	(7%)	0.0246
Visits count, mean (SD)	15.1 (15.5)		14.6 (16.8)		14.6 (13.9)		12.9 (11.1)		17.1 (17.4)		< 0.0001
Antihypertensive count, mean (SD)	2.3 (1.2)		1.6 (0.7)		1.5 (0.6)		1.5 (0.5)		3.8 (0.7)		< 0.0001

ADRD, Alzheimer's disease and related dementia; aTRH, apparent treatment‐resistant hypertension; HTN, hypertension; SD, standard deviation.

^a P‐value assesses the association across the HTN phenotypes.

^b Other races include American Indian or Alaska Native, Asian, Native Hawaiian or Pacific Islander, and Multiple Races.

Overall, 6401 (11.2%) patients developed ADRD, including 2539 (13%) aTRH patients, 1480 (10%) well‐controlled HTN patients, 1367 (11%) intermediate‐controlled HTN patients, and 1015 (9%) uncontrolled‐HTN patients (Table 1 ). The development of ADRD occurred significantly more often in the aTRH group compared to the other HTN control‐level groups (Table 1 ).

Variables associated with ADRD development

Table 2 shows variables that were significantly associated (P < 0.001) with the development of ADRD within the overall HTN population. In model 1, overall uncontrolled HTN was not significantly associated with the development of ADRD (Table 2 ). Model 2 included aTRH and HTN control levels and showed aTRH significantly associated with the development of ADRD (OR: 1.33, 95% CI: 1.23–1.43) (Table 2 ). When antihypertensive medications were included in model 3, aTRH was no longer among the significant variables, likely due to the collinearity with antihypertensive medications. However, model 3 showed that patients prescribed a thiazide diuretic (OR: 0.89, 95% CI: 0.84–0.96) or fixed‐dose combination medication (OR: 0.80, 95% CI: 0.73–0.88) had a lower risk of developing ADRD. In contrast, patients prescribed centrally acting α‐2 agonists were at higher risk of developing ADRD (OR: 1.27, 95% CI: 1.12–1.44) (Table 2 ).

Table 2.

Overall multivariate stepwise logistic regression for ADRD predictors in HTN population

Model 1a			Model 2b			Model 3c
Predictors	Odds ratio	95% CI	Predictors	Odds ratio	95% CI	Predictors	Odds ratio	95% CI
Age	1.05	(1.05–1.05)	aTRH	1.33	(1.23–1.43)	Age	1.05	(1.05–1.05)
Composite of anxiety and depression disorders	1.53	(1.44–1.62)	Intermediate‐controlled HTN	1.08	(1.00–1.17)	Composite of anxiety and depression disorders	1.45	(1.37–1.53)
Coronary artery disease	1.17	(1.10–1.25)	Uncontrolled HTN	0.96	(0.88–1.05)	Diabetes	1.12	(1.13–1.26)
Chronic kidney disease stage 4–5	1.32	(1.17–1.50)	Age	1.05	(1.05–1.05)	Stroke	1.40	(1.27–1.57)
Diabetes	1.27	(1.20–1.35)	Composite of anxiety and depression disorders	1.55	(1.47–1.64)	Alcohol‐related disorders	1.29	(1.15–1.45)
Sleep apnea	1.15	(1.06–1.24)	Coronary artery disease	1.15	(1.08–1.23)	Body mass index > 25 kg/m2	0.69	(0.65–0.73)
Stroke	1.44	(1.29–1.50)	Chronic kidney disease stage 4–5	1.26	(1.11–1.43)	Thiazide diuretics	0.89	(0.84–0.96)
Hypokalemia	1.21	(1.10–1.32)	Diabetes	1.24	(1.17–1.31)	Centrally acting α‐2 agonist	1.27	(1.12–1.44)
Alcohol‐related disorders	1.29	(1.15–1.45)	Stroke	1.42	(1.28–1.58)	Fixed‐dose combination	0.80	(0.73–0.88)
Body mass index > 25 kg/m2	0.69	(0.65–0.74)	Hypokalemia	1.19	(1.09–1.30)	Brain injury	1.32	(1.20–1.46)
Brain injury	1.35	(1.23–1.49)	Alcohol‐related disorders	1.29	(1.15–1.45)	SDI quartile 2 vs. 1	1.12	(1.03–1.21)
SDI quartile 2 vs. 1	1.14	(1.05–1.24)	Body mass index > 25 kg/m2	0.68	(0.64–0.73)	SDI quartile 3 vs. 1	1.16	(1.07–1.25)
SDI quartile 3 vs. 1	1.12	(1.11–1.29)	Brain injury	1.37	(1.24–1.51)	SDI quartile 4 vs. 1	1.17	(1.07–1.27)
SDI quartile 4 vs. 1	1.28	(1.18–1.40)	SDI quartile 2 vs. 1	1.13	(1.04–1.23)	Healthcare access quartile 2 vs. 1	0.67	(0.65–0.75)
Healthcare access quartile 2 vs. 1	0.68	(0.63–0.73)	SDI quartile 3 vs. 1	1.19	(1.10–1.28)	Healthcare access quartile 3 vs. 1	0.80	(0.74–0.87)
Healthcare access quartile 3 vs. 1	0.78	(0.72–0.85)	SDI quartile 4 vs. 1	1.24	(1.14–1.36)	Healthcare access quartile 4 vs. 1	0.69	(0.63–0.75)
Healthcare access quartile 4 vs. 1	0.65	(0.60–0.70)	Healthcare access quartile 2 vs. 1	0.66	(0.61–0.71)	Former smoking	1.13	(1.07–1.12)
Former smoking	1.17	(1.10–1.23)	Healthcare access quartile 3 vs. 1	0.77	(0.71–0.83)	Diastolic BP	1.00	(0.99–1.00)
Diastolic BP	1.00	(0.99–1.00)	Healthcare access quartile 4 vs. 1	0.64	(0.59–0.70)	Antihypertensive count	1.13	(1.10–1.16)
			Former smoking	1.17	(1.10–1.23)	Visit count	1.02	(1.02–1.02)
			Diastolic BP	1.00	(0.99–1.00)

ADRD, Alzheimer’s disease and related dementia; aTRH, apparent treatment‐resistant hypertension; BP, blood pressure; CI, confidence interval; HTN, hypertension; SDI, social deprivation index.

^a Model 1 included all the significant predictors from the univariate analysis plus the overall uncontrolled HTN (overall controlled HTN as a reference group).

^b Model 2 included all the significant predictors from the univariate analysis plus aTRH and other HTN control levels (well‐controlled HTN as a reference group).

^c Model 3 was using model 2 plus antihypertensive medications.

Notably, in all three models, we observed a linear relationship between SDI quartiles and the risk of ADRD; the higher the SDI, the greater the ADRD risk (Table 2 ). Conversely, low healthcare access (represented by quartiles 2, 3, and 4) was consistently associated with lower ADRD risk, compared to those with higher healthcare access across all three models (Table 2 ). Other consistent predictors across all models were age, the composite of anxiety and depression disorders, stroke, diabetes, brain injury, alcohol‐related disorders, and former smoking (Table 2 ). These main findings from the three models were mostly consistent in subgroup analyses of the overall controlled HTN cohort, overall uncontrolled HTN cohort, aTRH patients, patients of Black or African American race, Hispanic patients, patients with SDI < 50%, patients with healthcare access < 50%, and in each ADRD subtype (Tables S6 – S18 ).

The significant variables associated with ADRD development from the adjusted Cox proportional hazard model are presented in Table 3 . Overall, the results from this model were consistent with those from stepwise logistic regression analysis. When adjusting for HTN control levels in model 2, aTRH was significantly associated with increased risk of ADRD (HR: 1.16, 95% CI: 1.09–1.24, P < 0.0001), compared to well‐controlled HTN (Table 3 ). Neither intermediate‐controlled HTN nor uncontrolled HTN showed a significant association with ADRD risk in this model. Additionally, SDI and healthcare access showed similar patterns of association with ADRD risk as those observed in the stepwise logistic regression analysis (Table 3 ).

Table 3.

Multivariate cox proportional hazard model for ADRD predictors in HTN population ^a

Model 1b				Model 2c
Predictors	Hazard ratio	95% CI	P‐value	Predictors	Hazard ratio	95% CI	P‐value
Age	1.05	(1.05–1.05)	< 0.0001	aTRHd	1.16	(1.09–1.24)	< 0.0001
Brain injury	1.33	(1.22–1.45)	< 0.0001	Intermediate‐controlled HTNd	1.02	(0.94–1.01)	0.6628
Chronic kidney disease stage 4–5	1.31	(1.17–1.48)	< 0.0001	Uncontrolled HTNd	0.93	(0.86–1.01)	0.0729
Diabetes	1.25	(1.19–1.32)	< 0.0001	Age	1.05	(1.05–1.05)	< 0.0001
Dyslipidemia	0.89	(0.84–0.94)	< 0.0001	Composite of anxiety and depression disorders	1.51	(1.43–1.59)	< 0.0001
Composite of anxiety and depression disorders	1.50	(1.42–1.58)	< 0.0001	Hearing loss	1.08	(1.00–1.16)	0.0447
Heart failure with preserved ejection fraction	1.21	(1.11–1.31)	< 0.0001	Brain injury	1.34	(1.23–1.46)	< 0.0001
Stroke	1.44	(1.30–1.58)	< 0.0001	Chronic kidney disease stage 4–5	1.29	(1.15–1.46)	< 0.0001
Angina	1.10	(1.02–1.19)	0.0139	Diabetes	1.23	(1.17–1.30)	< 0.0001
Coronary artery disease	1.16	(1.08–1.24)	< 0.0001	Dyslipidemia	0.89	(0.840–0.94)	0.0001
Peripheral artery disease	1.15	(1.06–1.24)	0.0005	Heart failure with preserved ejection fraction	1.18	(1.09–1.28)	< 0.0001
Cardiomegaly	1.12	(1.03–1.22)	0.0094	Stroke	1.43	(1.30–1.58)	< 0.0001
Alcohol‐related disorders	1.40	(1.26–1.56)	< 0.0001	Angina	1.10	(1.02–1.19)	0.0143
Body mass index > 25 kg/m2	0.67	(0.63–0.71)	< 0.0001	Coronary artery disease	1.14	(1.07–1.22)	< 0.0001
SDI quartile 2 vs. 1	1.12	(1.04–1.21)	0.0039	Peripheral artery disease	1.14	(1.06–1.24)	0.0007
SDI quartile 3 vs. 1	1.14	(1.06–1.22)	0.0004	Cardiomegaly	1.11	(1.02–1.20)	0.0193
SDI quartile 4 vs. 1	1.17	(1.08–1.23)	0.0002	Alcohol‐related disorders	1.40	(1.26–1.56)	< 0.0001
Healthcare access quartile 2 vs. 1	0.86	(0.80–0.93)	< 0.0001	Body mass index > 25 kg/m2	0.66	(0.63–0.70)	< 0.0001
Healthcare access quartile 4 vs. 1	0.80	(0.74–0.86)	< 0.0001	SDI quartile 2 vs. 1	1.12	(1.04–1.21)	0.0036
Former smoking	1.13	(1.07–1.19)	< 0.0001	SDI quartile 3 vs. 1	1.14	(1.06–1.22)	0.0003
				SDI quartile 4 vs. 1	1.12	(1.07–1.27)	0.0004
				Healthcare access quartile 2 vs. 1	0.86	(0.80–0.92)	< 0.0001
				Healthcare access quartile 4 vs. 1	0.80	(0.74–0.86)	< 0.0001
				Former smoking	1.13	(1.07–1.19)	< 0.0001

Lancet risk factors: hypertension, diabetes, depression, brain injury, hearing loss, obesity, smoking, and alcohol‐related disorders. ADRD, Alzheimer’s disease and related dementia; aTRH, apparent treatment‐resistant hypertension; CI, confidence interval; HTN, hypertension; SDI, social deprivation index.

^a The table included only variables with significant P‐values except HTN control levels.

^b Model 1 adjusted for patient demographics, common comorbidities, and Lancet risk factors.

^c Model 2 adjusted for variables in model 1 plus aTRH and other HTN control levels.

^d The reference group is well‐controlled HTN.

ADRD outcomes stratified by HTN control levels

The cumulative hazard curves of ADRD events by aTRH and no aTRH, and by HTN control levels are shown in Figure 1 . aTRH was significantly associated with a higher cumulative hazard of ADRD over time, compared to no aTRH (P < 0.0001, Figure 1 a ). The median follow‐up times for aTRH, and no aTRH were similar (3.2 and 3.3 years, respectively). Similarly, when comparing aTRH with the HTN control levels, aTRH was significantly associated with a higher cumulative hazard of ADRD over time, compared to the other HTN control levels (P < 0.0001, Figure 1 b ). The median follow‐up times for aTRH, well‐controlled HTN, intermediate‐controlled HTN, and uncontrolled HTN were 3.2, 3.2, 3.3, and 3.1 years, respectively.

Figure 1.

Cumulative hazard curves of ADRD stratified by aTRH status and HTN control phenotypes. Panel (a) shows a significant difference in the cumulative hazard of ADRD between the groups (P < 0.0001), indicating a higher risk of ADRD among patients with aTRH. Panel (b) shows a significant difference among the HTN phenotypes (P < 0.0001), with the highest cumulative hazard of ADRD observed in patients with aTRH. In both panels, the analysis was conducted using the Nelson–Aalen estimator for cumulative hazard. The log‐rank test was applied to compare the cumulative hazard between the groups. The number‐at‐risk table below the graphs shows the number of participants at risk at various time points during the follow‐up. ADRD, Alzheimer’s disease and related dementia; aTRH, apparent treatment‐resistant hypertension; HTN, hypertension.

Subgroup analysis

The effect of aTRH on ADRD risk in subgroups is presented in Figure 2 . aTRH was significantly associated with increased ADRD risk across the majority of subgroups, compared to those with no aTRH. These results were consistent with the results from the stepwise logistic regression modeling (Table 2 ), and other subgroup analyses (Tables S6 – S18 ).

Figure 2.

aTRH risk for ADRD in subgroups. This forest plot presents the adjusted odds ratios and 95% confidence intervals for the association between aTRH and ADRD across various subgroups. The analysis was conducted using multivariate logistic regression, adjusting for patient demographics (age, sex, race, and ethnicity) and common comorbidities (diabetes, stroke, CKD, brain injury, and composite of anxiety and depression disorders). ADRD, Alzheimer’s disease and related dementia; aTRH, apparent treatment‐resistant hypertension; CKD, chronic kidney disease; CI, confidence interval; HTN, hypertension; SDI, social deprivation index.

DISCUSSION

Previous studies have shown that HTN is associated with a higher risk of ADRD; however, these studies often used a single BP measurement or HTN diagnosis to categorize patients and did not consider variation in HTN and BP over time. ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ In our study, we addressed this gap by categorizing HTN into aTRH and varying HTN control levels, and examining their impact on ADRD risk. We found that aTRH was associated with increased risk of ADRD. Our models also validated the majority of the risk factors identified in the 2020 Lancet Commission on Dementia Prevention, Intervention, and Care. ⁵ Of importance, our findings demonstrated that the higher the SDI, the greater the ADRD risk. These findings were consistent across subgroups, and to our knowledge, this is the first study highlighting the differential impact of aTRH and varying HTN control levels on ADRD risk.

Our results illustrate that aTRH is significantly associated with an increased risk of ADRD. This finding advances our understanding of HTN as a modifiable risk factor for ADRD as it suggests that HTN control levels contribute to cognitive impairment differently. The risk associated with aTRH was consistent across ADRD phenotypes and not just limited to vascular dementia. Our findings extend the results from the Lifestyle Modification for Resistant Hypertension (TRIUMPH) trial, which indicated that neurocognitive impairment is common among patients with resistant HTN, with more than one‐third of the participants exhibiting poor performance on cognitive function tests. ¹⁷ Additionally, our results are in line with previous studies, which revealed that uncontrolled HTN is associated with a greater risk of ADRD, compared to controlled HTN. However, unlike these studies, our study characterized HTN phenotypes using multiple BP measurements, and the number of antihypertensive classes prescribed longitudinally. ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ This distinction demonstrated in our study between HTN control levels and ADRD risk is crucial as it implies the importance of intensive BP control and targeted interventions to mitigate future ADRD risk in aTRH patients.

Failure to achieve BP targets in aTRH exacerbates neurodegenerative pathways through several mechanisms. ³³ Sustained HTN damages cerebral blood vessels, leading to structural and functional cerebrovascular changes such as white matter lesions, micro‐infarcts, and microbleeds, all of which are known to cause ADRD. ³⁴ Chronic high BP compromises the blood–brain barrier (BBB), increasing vulnerability to neurotoxic substances like beta‐amyloid, and ultimately leading to inflammation, neuronal damage, and cognitive decline. ³⁵ Furthermore, chronic HTN reduces cerebral blood flow over time, causing brain atrophy, particularly in areas like the hippocampus that are essential for memory. ³⁶ HTN increases the phosphorylation of tau protein, contributing to neurofibrillary tangle formation, a key component of Alzheimer’s pathology. ³⁷ In addition to these vascular and neurodegenerative effects, individuals with aTRH frequently present with a high burden of comorbidities, including diabetes, chronic kidney disease, coronary artery disease, and heart failure. ²² , ³⁸ In our cohort, many of these conditions were independently associated with ADRD, highlighting a multifactorial risk profile in which both uncontrolled BP burden and coexisting cardiovascular and metabolic diseases contribute to dementia pathogenesis. ³⁹ , ⁴⁰ Collectively, these mechanisms highlight the profound impact of uncontrolled HTN and its associated comorbidities on cerebrovascular health and cognitive function, emphasizing the importance of effective BP management and comprehensive cardiovascular risk reduction in mitigating dementia risk.

Interestingly, dyslipidemia was associated with a lower risk of ADRD in our multivariable model. This likely reflects the protective effect of co‐prescribed cardiovascular therapies, like lipid‐lowering agents and antithrombotic medications, rather than a true protective effect of dyslipidemia itself. These therapies have been associated with improved vascular health, reduced inflammation, and better cerebral perfusion, which may help mitigate cognitive decline. ⁴¹ , ⁴²

A recent meta‐analysis of longitudinal cohort studies showed that treatment with antihypertensive medications was associated with a 21% reduction in dementia risk, compared to untreated HTN. ¹² Moreover, emerging evidence from observational studies suggests that some classes of antihypertensive medications may offer greater reductions in dementia risk than others, independently or through their BP‐lowering effect. ⁴³ , ⁴⁴ In the context of randomized clinical trials, the results on the beneficial effect of antihypertensive medications on ADRD risk have been inconsistent. Both the Systolic Blood Pressure Intervention Trial Memory and Cognition in Decreased Hypertension (SPRINT‐MIND) and Systolic Hypertension in Europe (Syst‐Eur) studies showed that antihypertensive treatment was protective against ADRD risk. ⁴⁵ , ⁴⁶ So far, no clinical trial has primarily evaluated the effect of specific antihypertensive classes on ADRD outcomes. Our results in this regard, are in line with evidence from observational studies, showing that patients prescribed centrally acting α‐2 agonist had a higher risk of ADRD, whereas patients prescribed thiazide diuretics and fixed‐dose combination medications had a lower risk of developing ADRD.

The protective effect of thiazide diuretics in this study supports the results from the Cache County Study, which found that thiazide diuretics were associated with a 30% reduction in ADRD risk. ⁴⁷ This protective effect is likely due to the anti‐amyloidogenic, anti‐inflammatory, and anti‐thromboembolic properties of thiazide diuretics. ⁴⁸ The reduced ADRD risk associated with fixed‐dose combination medications confirms the findings of Verma et al., which demonstrated better clinical outcomes with fixed‐dose combinations compared to multipill therapy. ⁴⁹ This favorable effect of fixed‐dose combination therapy could be explained by the better adherence, synergistic pharmacological actions, fewer side effects, and lower cost of fixed‐dose combination medications, contributing to better HTN control and lower ADRD risk. ⁵⁰

Our study further illustrates the role of SDI in ADRD risk, with a higher SDI correlated with increased ADRD risk, suggesting a health disparity. This relationship may be driven by limited healthcare access, poor nutrition, physical inactivity, exposure to stress, and low education, which negatively impact cognitive function. ⁵¹ Our results align with previous findings linking a higher SDI with ADRD. ⁵² This finding suggests that enhancing social health and support can effectively reduce health disparities and ADRD risk, beyond medical treatment. Low healthcare access was associated with lower ADRD risk in our study, possibly due to ADRD underdiagnosis. Indeed, evidence suggests that with limited healthcare access, many ADRD cases do not get the necessary medical evaluations to be identified and could go undiagnosed and unrecorded. ⁵³

Previous studies indicated that Black or African Americans with uncontrolled HTN are at greater risk for ADRD, compared to other races. ⁵⁴ , ⁵⁵ In our primary analysis, Black or African American race was not significantly associated with the development of ADRD; however, Black or African American race was significantly associated with the development of some of the individual ADRD phenotypes. Additionally, in subgroup analyses, we revealed a compelling linear relationship among the HTN control levels in the Black or African American race group (Table S9 ). These results may suggest that the higher ADRD risk in this population may be driven by the degree of BP control. This relationship may also explain how variance in BP control contributes to racial disparities in ADRD risk. ⁵

Our study is not without limitations. First, many ADRD cases go undiagnosed, ⁵³ which could potentially lead to an underestimation of the prevalence and risk factors associated with ADRD. However, our study was still able to observe incidence rates similar to those previously reported. ⁵⁶ Second, the use of antihypertensive medications for the treatment of heart failure might have led to misclassification of some aTRH patients. However, excluding heart failure patients would also impact the generalizability, as heart failure is common in the HTN population. Third, we did not assess medication adherence, but relied on antihypertensive prescriptions, which may not reflect actual treatment effectiveness. However, we used validated computable phenotypes, which required included patients to receive routine outpatient care, have multiple encounters, and meet the phenotype criteria at least twice to improve the quality of our study. Finally, SDI was derived from the most recent residential ZIP code available in the EHR, which may not fully capture changes in neighborhood socioeconomic status over time. This approach might have introduced exposure misclassification for patients who moved during the study period.

Despite these limitations, our study leveraged a large, longitudinal, and racially diverse patient population representative of the Floridian and U.S. populations, enhancing the generalizability of our findings. Our study contributes to existing literature by elucidating the effect of differential and granular levels of HTN control on ADRD risk. The broad inclusion of variables, such as HTN control levels, antihypertensive medications, SDI, healthcare access levels, and comorbid conditions, provides a holistic understanding of ADRD risk factors. Our validation of 2020 Lancet ADRD risk factors further reinforces the credibility of our findings.

In conclusion, our study provides significant insights into variables associated with ADRD development within HTN patients. We found that a more complex HTN phenotype like aTRH was significantly associated with the development of ADRD. We also showed that patients with high SDI (lower socioeconomic status) were at higher risk of ADRD development. Our study highlights the importance of comprehensive HTN control and socioeconomic interventions in preventing or reducing ADRD. These findings suggest that clinicians should prioritize intensive management of aTRH and that public health initiatives should focus on improving healthcare access and socioeconomic conditions in deprived communities to mitigate ADRD risk.

FUNDING

This project was supported by a Development Grant from the 1Florida Alzheimer’s Disease Research Center and NIH P30 AG066506. CWM was supported by NIH grants K01 HL141690, R03 HL172123, and R03 HL172987. SMS was supported by NIH grant AG066506. Research reported in this publication was supported in part by the OneFlorida+ Clinical Research Network, funded by the Patient‐Centered Outcomes Research Institute numbers CDRN‐1501‐26692, RI‐CRN‐2020‐005, and RI‐FLORIDA‐01‐PS1, and in part by the University of Florida Clinical and Translational Science Institute, which is supported in part by the NIH National Center for Advancing Translational Sciences under award numbers UL1TR001427 and UL1TR000064. The content is solely the responsibility of the authors and does not necessarily represent the official views of the Patient‐Centered Outcomes Research Institute (PCORI), its Board of Governors or Methodology, the OneFlorida+ Clinical Research Network, the UF‐FSU Clinical and Translational Science Institute, or the National Institutes of Health.

CONFLICT OF INTEREST

The authors declared no competing interests for this work.

AUTHOR CONTRIBUTIONS

E.A.J., M.A., and C.W.M. wrote the manuscript; E.A.J., S.M.S., Y.G., G.E.S., and C.W.M. designed the research; E.A.J., M.A.Z., and C.W.M. performed the research; E.A.J., M.A., M.A.Z., and C.W.M. analyzed the data.

Supporting information

Data S1.

DATA AVAILABILITY STATEMENT

The data underlying this article cannot be shared publicly as the datasets include HIPAA‐limited data. Qualified investigators who wish to utilize data from the OneFlorida Data Trust can apply through their respective Front Door policies and procedures. The underlying computable phenotypes and analysis codes will be shared on reasonable request to the corresponding author.